Jan 2023 – Jun 2023 · 5 mos · New York, New York, United States · On-site

• ● Owned end-to-end development & productionization of a multilingual machine translation system by finetuning Meta’s M2M-100 model that could translate workforce data (such as job titles) from 23 languages to English; created a custom loss function using bertscore; the final system achieved a BLEU score of 50+ for all languages
• ● Incorporated the model into daily data ingestion pipeline using Amazon EKS, Jenkins, and Github Actions; the model translates ~3 million+ titles daily
• ● Corrected representativeness of data using cross sectional scaling - Used labor data to approximate scaling factors to adjust counts of different jobs (SOC codes) across states and industries (NAICS) in the USA; applied exponential smoothing based on FastText embeddings of jobs and their cosine distance from each other while correcting for the counts

Sep 2022 – Dec 2022 · 3 mos

• Graduate Teaching Assistant for Artificial Intelligence (COMS W4701)

Mar 2022 – Dec 2022 · 9 mos

• Supported the Data Science Institute in organizing academic events such as research fair, poster sessions, career fairs and more for the graduate and undergraduate student community at Columbia

Feb 2022 – May 2022 · 3 mos

• Wrote python scripts to parse and clean LinkedIn profiles to build a dataset of mental health workforce in NY; employed Named Entity Recognition models to extract information such as school name, firm name, and location out of raw LinkedIn profiles
• Provided data-based evidence to showcase the impact of policy changes made in the year 2010 by the New York State Office of Mental Health on the mental health workforce (chiefly how contingent workforce expanded and the quality of care provided might have been affected)

Sep 2022 – Dec 2022 · 3 mos · New York, United States · Remote

• ● Worked on benchmarking publicly available SOTA models for triple extraction (REBEL, AllenNLP, & Stanford OIE) while building a pipeline for building knowledge graph based question answering system for financial research papers
• ● Completely owned research & development of building a natural language question to SPARQL query translation system for fetching information out of the built knowledge graph - broke down the system into 4 components: query template classification model (via distilBERT), entities & relation extraction model (via dependency parsing), query triple to ontology triple matching (using finBERT embedding similarity & levenshtein distance), and SPARQL query generation; this system had an efficacy rate of ~85%

May 2022 – Aug 2022 · 3 mos · New York, New York, United States

• Developed a multilingual machine translation system finetuned to workforce domain using Meta’s M2M-100 transformer model
• Applied dynamic quantization to the model during production improving the inference speed by ~3x; further built a jenkins pipeline that automates monthly prediction refresh at the click of a button
• Explored state-of-the-art language models like M2M_100, MarianMT, mT5, and multilingual BERT
• Also built a fastText based language detection model that outperforms google translate
• while detecting 24 languages

Apr 2021 – Jun 2021 · 2 mos

• Focused on commercial due diligence and pre-DD support (e.g. M&A target screening and assessment, market POV, competitor benchmarking)
• Executed multi-country surveys to evaluate customer insights and analyze strategic initiatives including innovation, pricing and distribution of portfolio products
• Conducted disruption assessments to understand the level of digital disruption in the market and the impact to incumbents
• Employed K-modes clustering and performed customer segmentation for studying customer behavior of 1,500+ survey participants while analyzing the QSR market in the Indian subcontinent
• Coached fellow Bainees on the proprietary toolkit

Jul 2020 – Mar 2021 · 8 mos

• Collaborated with business leadership at Baincubator (Bain’s corporate incubator) while conceptualizing a new engine 2 business for Bain; spearheaded market research as well as development of multiple ML based POCs for core product
• Utilized DeepPavlov’s Open Domain Question Answering model to build an efficient knowledge sharing solution for corporates; the solution was declared winner of Bain’s Global Innovation Hackathon, 2020
• Provided analytical support while contributing to the development of Asia Pacific PE Report (2021)

Jun 2019 – Jun 2020 · 1 yr

• Led development & deployment of various machine-learning enabled IP tools for the Bain PEG ecosystem in a bid to automate data-driven analysis thereby reducing turnaround time for various processes by ~90%. Key projects discussed below:
• (i) Collaborated in a team of three and developed an automated survey data analysis and visualization tool using Alteryx, Tableau and Python
• (ii) Built a machine learning powered SEO analytics engine that takes in online visibility data for different brands from Semrush and SimilarWeb and churns out key benchmarking metrics such as SEO positioning, indexed web traffic, keyword search performance etc.
• (iii) Wrote python scripts for frequent text analysis tasks such as unaided brand awareness analysis, job title standardization for workforce analytics and more

Jun 2018 – Jul 2018 · 1 mo · Gurgaon, India

• Experimented with various statistical (GMM-HMM), machine learning (logistic regression, SVM), and deep learning (DNN, CNN) models as well as different representations for audio signals while building a Speech Emotion Recognition system
• Achieved best in class accuracy (~85%) while using DNN architecture on audio signals compressed using PCA
• Proposed and developed an automated short text tagging tool using ​LDA for an international pharma client​

Dec 2016 – May 2017 · 5 mos · New Delhi Area, India

• Among the 43 candidates out of 1000+ selected for the internship to participate in day to day activities of university's administration.
• Worked with the Dean PG office to digitalize the records of post graduate students in DTU